Weights

Tom DiBenedetto tdib at UMICH.EDU
Tue Mar 4 15:26:25 CST 1997


On Sun, 02 Mar 1997 10:35:36 -0800, hoelzer at med.unr.edu wrote:

> As you seem to agree that any
>weighting scheme is necessarily arbitrary (although I would argue that some are
>more justifiable than others), there is no a priori reason to choose equal
>weighting.  Equal weighting requires as much justification as any other
>weighting scheme because it makes assertions about the past history of
>character-state changes in your group.

No it doesnt. YOU are asserting that the "past history of
character-state changes in my group" are a relevant factor in making
the decision as to whether this or that single, particular event
happened or not. I do not consider that factor to be relevant, and my
equal weighting thus does not even consider such factors, and makes
no assertions whatsoever on that score. It is justified for very
different reason, according to the standards of my own conceptual
approach. See other postings.

  I see no justification of equal
>weighting as a default assumption in the absence of any information.  I would
>also argue that it particular weighting schemes can be justified based on
>information taken from independent evidence (e.g., transistion/transversion
>ratios).  We can get a sense of the generality of such patterns when they appear
>over and over again in independent data sets.

I agree that we can get a sense of the generality of such matters,
and
that such generalities are valuable scientific knowledge. I do not
consider them to be very helpful in terms of determining what did or
did not occur in specific cases.

  I grant you that you are always
>making a leap of faith in generalizing such patterns to a new group, but at
>least it is a leap based on explicit criteria.  You also take a leap of faith
>with equal weighting (i.e., you are asserting as much knowledge about past
>processes with equal weighting as you are with any other weighting scheme).  Can
>you please outline your justification of equal weighting as a default weighting
>scheme that makes fewer assumptions about past processes than does any other
>weighting scheme?

No, because past processes are simply not a relevant consideration.
They refer to other situations. Even if my particular set of
transformations are consistent with the generalizations, they are so
on a general level,,,i.e. my transition/transversion ratios might be
consistent with what has been found in other studies, but that does
not really help me determine if this specific transition occurred or
not.

>>......your a priori imposition of a measured amount of relevance
>>to a particular transformation leads to a phylogeny based on what you
>>think you already know, rather than one which is structured to allow
>>discovery of new facts.
>
>I disagree.  First, to the extent that this statement is true it certainly
>applies to the case of equal weighting, as well as any other weighting scheme.
>Just because you may choose equal weighting in an effort to accomodate an
>admitted lack of historical information for your group (a weakly justifiable
>position in my view) does not mean that your weighting scheme will not mislead
>you.

My weighting scheme will mislead me to the extent that some of the
things I weight 1, should be weighed 0, and some I weigh zero (by not
including) are real (i.e. 1),,,although I doubt that there are many
instances of the latter.  As we approach the truth, we should
accumulate a set of real transformations for any group (a set of
one-weighted characters) and a set of zeros lying on the side. My
method of closing in on that is to filter my characters through a
series of tests which convert some subset of 1s to 0s according the
expectations we all  have for homologies. You are assigning
intermediate values to all trasfomations on the basis of process
considerations which I consider to have weak predictive power in
particular cases, thus not appopriate for inclusion in the
tree-finding step.

>Second, while any arbitrary weighting scheme (including equal weighting)
>can influence your phylogenetic conclusions, most forms of phylogenetic analyses
>still permit the "discovery of new facts" (I would prefer to say the "production
>of new phylogenetic hypotheses").

But those which emerge from a weighted analysis based on process
"knowledge" reflect your efforts to conform the results to the
generalized expectations you derive
from elsewhere. This is a form of verificationism; your trees will,
to some extent, reflect the assumptions you pump into the process;
they will serve to verify previous knowledge rather than challange
it. Your process generalizations are hypotheses in their own right,
and they should be tested by the new phylogeny you reconstruct. You
are not allowing that to happen by building your process hypothesis
into the procedure for finding new trees. You are immunizing your
process hypotheses from falsification. Bad move.




More information about the Taxacom mailing list